Start With the Riskiest Assumption, Not the Coolest Agent
Maurya's core move is to rank your assumptions by how fatal they would be if wrong, then attack the most dangerous one first. For an AI-native founder, that ranking almost always surfaces something about the agent before it surfaces something about the product vision. Can the agent actually complete the task reliably? Does your eval harness even exist yet? If you cannot measure output quality, you cannot trust the system in production, and everything built on top of it is speculation dressed as a product.
Problem Interviews Apply to the Workflow, Not Just the Customer
Maurya separates problem interviews from solution interviews because understanding pain comes before validating the fix. The same logic holds for agentic design. Before you wire up tools and retrieval, interview the workflow: where does it branch unpredictably, where does it require context a model cannot reliably hold, where does a wrong output compound into something unrecoverable? Most problems that look like agent problems are actually workflow problems — fixed paths that need better scoping, not more autonomy. Solve the workflow first, then promote to agency only where genuine branching earns it.
The Lean Canvas Is an Eval in Disguise
The lean canvas forces you to write down your unfair advantage, your customer segments, and your cost structure in one view so the contradictions become visible. An eval harness does the same thing for an agentic system: it makes the failure modes legible before they reach a user. Maurya insists you build the canvas before you build the product; the parallel discipline is building the eval before you build the agent. Both artifacts are thinking tools that surface what you are quietly assuming is true.
Human Judgment Is the Stage Gate Between Iterations
Running Lean is iterative by design — you run an experiment, measure, learn, and decide whether to persevere or pivot. That decision point requires a human making a judgment call about evidence, not a system optimizing a metric. In an AI-native company, the same gate applies inside the product: agents handle the repeatable work, but a human holds accountability for the tradeoffs, the trust boundaries, and the call to ship or stop. Maurya's cadence assumes a founder willing to face bad news early. That posture is exactly what keeps autonomous systems from drifting past their intended scope.
